Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * An async IO implementation for Linux |
| 3 | * Written by Benjamin LaHaise <bcrl@kvack.org> |
| 4 | * |
| 5 | * Implements an efficient asynchronous io interface. |
| 6 | * |
| 7 | * Copyright 2000, 2001, 2002 Red Hat, Inc. All Rights Reserved. |
| 8 | * |
| 9 | * See ../COPYING for licensing terms. |
| 10 | */ |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 11 | #define pr_fmt(fmt) "%s: " fmt, __func__ |
| 12 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 13 | #include <linux/kernel.h> |
| 14 | #include <linux/init.h> |
| 15 | #include <linux/errno.h> |
| 16 | #include <linux/time.h> |
| 17 | #include <linux/aio_abi.h> |
Paul Gortmaker | 630d9c4 | 2011-11-16 23:57:37 -0500 | [diff] [blame] | 18 | #include <linux/export.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 19 | #include <linux/syscalls.h> |
Jens Axboe | b9d128f | 2009-10-29 13:59:26 +0100 | [diff] [blame] | 20 | #include <linux/backing-dev.h> |
Badari Pulavarty | 027445c | 2006-09-30 23:28:46 -0700 | [diff] [blame] | 21 | #include <linux/uio.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 22 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 23 | #include <linux/sched.h> |
| 24 | #include <linux/fs.h> |
| 25 | #include <linux/file.h> |
| 26 | #include <linux/mm.h> |
| 27 | #include <linux/mman.h> |
Michael S. Tsirkin | 3d2d827 | 2009-09-21 17:03:51 -0700 | [diff] [blame] | 28 | #include <linux/mmu_context.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 29 | #include <linux/slab.h> |
| 30 | #include <linux/timer.h> |
| 31 | #include <linux/aio.h> |
| 32 | #include <linux/highmem.h> |
| 33 | #include <linux/workqueue.h> |
| 34 | #include <linux/security.h> |
Davide Libenzi | 9c3060b | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 35 | #include <linux/eventfd.h> |
Jeff Moyer | cfb1e33 | 2009-10-02 18:57:36 -0400 | [diff] [blame] | 36 | #include <linux/blkdev.h> |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 37 | #include <linux/compat.h> |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 38 | #include <linux/anon_inodes.h> |
| 39 | #include <linux/migrate.h> |
| 40 | #include <linux/ramfs.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 41 | |
| 42 | #include <asm/kmap_types.h> |
| 43 | #include <asm/uaccess.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 44 | |
Al Viro | 68d70d0 | 2013-06-19 15:26:04 +0400 | [diff] [blame] | 45 | #include "internal.h" |
| 46 | |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 47 | #define AIO_RING_MAGIC 0xa10a10a1 |
| 48 | #define AIO_RING_COMPAT_FEATURES 1 |
| 49 | #define AIO_RING_INCOMPAT_FEATURES 0 |
| 50 | struct aio_ring { |
| 51 | unsigned id; /* kernel internal index number */ |
| 52 | unsigned nr; /* number of io_events */ |
| 53 | unsigned head; |
| 54 | unsigned tail; |
| 55 | |
| 56 | unsigned magic; |
| 57 | unsigned compat_features; |
| 58 | unsigned incompat_features; |
| 59 | unsigned header_length; /* size of aio_ring */ |
| 60 | |
| 61 | |
| 62 | struct io_event io_events[0]; |
| 63 | }; /* 128 bytes + ring size */ |
| 64 | |
| 65 | #define AIO_RING_PAGES 8 |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 66 | |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 67 | struct kioctx { |
| 68 | atomic_t users; |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 69 | atomic_t dead; |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 70 | |
| 71 | /* This needs improving */ |
| 72 | unsigned long user_id; |
| 73 | struct hlist_node list; |
| 74 | |
Kent Overstreet | 3e845ce | 2013-05-07 16:18:51 -0700 | [diff] [blame] | 75 | /* |
| 76 | * This is what userspace passed to io_setup(), it's not used for |
| 77 | * anything but counting against the global max_reqs quota. |
| 78 | * |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 79 | * The real limit is nr_events - 1, which will be larger (see |
Kent Overstreet | 3e845ce | 2013-05-07 16:18:51 -0700 | [diff] [blame] | 80 | * aio_setup_ring()) |
| 81 | */ |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 82 | unsigned max_reqs; |
| 83 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 84 | /* Size of ringbuffer, in units of struct io_event */ |
| 85 | unsigned nr_events; |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 86 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 87 | unsigned long mmap_base; |
| 88 | unsigned long mmap_size; |
| 89 | |
| 90 | struct page **ring_pages; |
| 91 | long nr_pages; |
| 92 | |
Kent Overstreet | 4e23bca | 2013-05-07 16:18:56 -0700 | [diff] [blame] | 93 | struct rcu_head rcu_head; |
| 94 | struct work_struct rcu_work; |
| 95 | |
| 96 | struct { |
| 97 | atomic_t reqs_active; |
| 98 | } ____cacheline_aligned_in_smp; |
| 99 | |
| 100 | struct { |
| 101 | spinlock_t ctx_lock; |
| 102 | struct list_head active_reqs; /* used for cancellation */ |
| 103 | } ____cacheline_aligned_in_smp; |
| 104 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 105 | struct { |
| 106 | struct mutex ring_lock; |
Kent Overstreet | 4e23bca | 2013-05-07 16:18:56 -0700 | [diff] [blame] | 107 | wait_queue_head_t wait; |
| 108 | } ____cacheline_aligned_in_smp; |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 109 | |
| 110 | struct { |
| 111 | unsigned tail; |
| 112 | spinlock_t completion_lock; |
Kent Overstreet | 4e23bca | 2013-05-07 16:18:56 -0700 | [diff] [blame] | 113 | } ____cacheline_aligned_in_smp; |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 114 | |
| 115 | struct page *internal_pages[AIO_RING_PAGES]; |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 116 | struct file *aio_ring_file; |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 117 | }; |
| 118 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 119 | /*------ sysctl variables----*/ |
Zach Brown | d55b5fd | 2005-11-07 00:59:31 -0800 | [diff] [blame] | 120 | static DEFINE_SPINLOCK(aio_nr_lock); |
| 121 | unsigned long aio_nr; /* current system wide number of aio requests */ |
| 122 | unsigned long aio_max_nr = 0x10000; /* system wide maximum number of aio requests */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 123 | /*----end sysctl variables---*/ |
| 124 | |
Christoph Lameter | e18b890 | 2006-12-06 20:33:20 -0800 | [diff] [blame] | 125 | static struct kmem_cache *kiocb_cachep; |
| 126 | static struct kmem_cache *kioctx_cachep; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 127 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 128 | /* aio_setup |
| 129 | * Creates the slab caches used by the aio routines, panic on |
| 130 | * failure as this is done early during the boot sequence. |
| 131 | */ |
| 132 | static int __init aio_setup(void) |
| 133 | { |
Christoph Lameter | 0a31bd5 | 2007-05-06 14:49:57 -0700 | [diff] [blame] | 134 | kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
| 135 | kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 136 | |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 137 | pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 138 | |
| 139 | return 0; |
| 140 | } |
H Hartley Sweeten | 385773e | 2009-09-22 16:43:53 -0700 | [diff] [blame] | 141 | __initcall(aio_setup); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 142 | |
| 143 | static void aio_free_ring(struct kioctx *ctx) |
| 144 | { |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 145 | int i; |
| 146 | struct file *aio_ring_file = ctx->aio_ring_file; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 147 | |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 148 | for (i = 0; i < ctx->nr_pages; i++) { |
| 149 | pr_debug("pid(%d) [%d] page->count=%d\n", current->pid, i, |
| 150 | page_count(ctx->ring_pages[i])); |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 151 | put_page(ctx->ring_pages[i]); |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 152 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 153 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 154 | if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages) |
| 155 | kfree(ctx->ring_pages); |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 156 | |
| 157 | if (aio_ring_file) { |
| 158 | truncate_setsize(aio_ring_file->f_inode, 0); |
| 159 | pr_debug("pid(%d) i_nlink=%u d_count=%d d_unhashed=%d i_count=%d\n", |
| 160 | current->pid, aio_ring_file->f_inode->i_nlink, |
| 161 | aio_ring_file->f_path.dentry->d_count, |
| 162 | d_unhashed(aio_ring_file->f_path.dentry), |
| 163 | atomic_read(&aio_ring_file->f_inode->i_count)); |
| 164 | fput(aio_ring_file); |
| 165 | ctx->aio_ring_file = NULL; |
| 166 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 167 | } |
| 168 | |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 169 | static int aio_ring_mmap(struct file *file, struct vm_area_struct *vma) |
| 170 | { |
| 171 | vma->vm_ops = &generic_file_vm_ops; |
| 172 | return 0; |
| 173 | } |
| 174 | |
| 175 | static const struct file_operations aio_ring_fops = { |
| 176 | .mmap = aio_ring_mmap, |
| 177 | }; |
| 178 | |
| 179 | static int aio_set_page_dirty(struct page *page) |
| 180 | { |
| 181 | return 0; |
| 182 | } |
| 183 | |
Benjamin LaHaise | 0c45355 | 2013-07-17 09:34:24 -0400 | [diff] [blame^] | 184 | #if IS_ENABLED(CONFIG_MIGRATION) |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 185 | static int aio_migratepage(struct address_space *mapping, struct page *new, |
| 186 | struct page *old, enum migrate_mode mode) |
| 187 | { |
| 188 | struct kioctx *ctx = mapping->private_data; |
| 189 | unsigned long flags; |
| 190 | unsigned idx = old->index; |
| 191 | int rc; |
| 192 | |
| 193 | /* Writeback must be complete */ |
| 194 | BUG_ON(PageWriteback(old)); |
| 195 | put_page(old); |
| 196 | |
| 197 | rc = migrate_page_move_mapping(mapping, new, old, NULL, mode); |
| 198 | if (rc != MIGRATEPAGE_SUCCESS) { |
| 199 | get_page(old); |
| 200 | return rc; |
| 201 | } |
| 202 | |
| 203 | get_page(new); |
| 204 | |
| 205 | spin_lock_irqsave(&ctx->completion_lock, flags); |
| 206 | migrate_page_copy(new, old); |
| 207 | ctx->ring_pages[idx] = new; |
| 208 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
| 209 | |
| 210 | return rc; |
| 211 | } |
Benjamin LaHaise | 0c45355 | 2013-07-17 09:34:24 -0400 | [diff] [blame^] | 212 | #endif |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 213 | |
| 214 | static const struct address_space_operations aio_ctx_aops = { |
| 215 | .set_page_dirty = aio_set_page_dirty, |
Benjamin LaHaise | 0c45355 | 2013-07-17 09:34:24 -0400 | [diff] [blame^] | 216 | #if IS_ENABLED(CONFIG_MIGRATION) |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 217 | .migratepage = aio_migratepage, |
Benjamin LaHaise | 0c45355 | 2013-07-17 09:34:24 -0400 | [diff] [blame^] | 218 | #endif |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 219 | }; |
| 220 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 221 | static int aio_setup_ring(struct kioctx *ctx) |
| 222 | { |
| 223 | struct aio_ring *ring; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 224 | unsigned nr_events = ctx->max_reqs; |
Zach Brown | 41003a7 | 2013-05-07 16:18:25 -0700 | [diff] [blame] | 225 | struct mm_struct *mm = current->mm; |
Michel Lespinasse | 41badc1 | 2013-02-22 16:32:47 -0800 | [diff] [blame] | 226 | unsigned long size, populate; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 227 | int nr_pages; |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 228 | int i; |
| 229 | struct file *file; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 230 | |
| 231 | /* Compensate for the ring buffer's head/tail overlap entry */ |
| 232 | nr_events += 2; /* 1 is required, 2 for good luck */ |
| 233 | |
| 234 | size = sizeof(struct aio_ring); |
| 235 | size += sizeof(struct io_event) * nr_events; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 236 | |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 237 | nr_pages = PFN_UP(size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 238 | if (nr_pages < 0) |
| 239 | return -EINVAL; |
| 240 | |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 241 | file = anon_inode_getfile_private("[aio]", &aio_ring_fops, ctx, O_RDWR); |
| 242 | if (IS_ERR(file)) { |
| 243 | ctx->aio_ring_file = NULL; |
| 244 | return -EAGAIN; |
| 245 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 246 | |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 247 | file->f_inode->i_mapping->a_ops = &aio_ctx_aops; |
| 248 | file->f_inode->i_mapping->private_data = ctx; |
| 249 | file->f_inode->i_size = PAGE_SIZE * (loff_t)nr_pages; |
| 250 | |
| 251 | for (i = 0; i < nr_pages; i++) { |
| 252 | struct page *page; |
| 253 | page = find_or_create_page(file->f_inode->i_mapping, |
| 254 | i, GFP_HIGHUSER | __GFP_ZERO); |
| 255 | if (!page) |
| 256 | break; |
| 257 | pr_debug("pid(%d) page[%d]->count=%d\n", |
| 258 | current->pid, i, page_count(page)); |
| 259 | SetPageUptodate(page); |
| 260 | SetPageDirty(page); |
| 261 | unlock_page(page); |
| 262 | } |
| 263 | ctx->aio_ring_file = file; |
| 264 | nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) |
| 265 | / sizeof(struct io_event); |
| 266 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 267 | ctx->ring_pages = ctx->internal_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 268 | if (nr_pages > AIO_RING_PAGES) { |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 269 | ctx->ring_pages = kcalloc(nr_pages, sizeof(struct page *), |
| 270 | GFP_KERNEL); |
| 271 | if (!ctx->ring_pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 272 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 273 | } |
| 274 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 275 | ctx->mmap_size = nr_pages * PAGE_SIZE; |
| 276 | pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size); |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 277 | |
Zach Brown | 41003a7 | 2013-05-07 16:18:25 -0700 | [diff] [blame] | 278 | down_write(&mm->mmap_sem); |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 279 | ctx->mmap_base = do_mmap_pgoff(ctx->aio_ring_file, 0, ctx->mmap_size, |
| 280 | PROT_READ | PROT_WRITE, |
| 281 | MAP_SHARED | MAP_POPULATE, 0, &populate); |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 282 | if (IS_ERR((void *)ctx->mmap_base)) { |
Zach Brown | 41003a7 | 2013-05-07 16:18:25 -0700 | [diff] [blame] | 283 | up_write(&mm->mmap_sem); |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 284 | ctx->mmap_size = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 285 | aio_free_ring(ctx); |
| 286 | return -EAGAIN; |
| 287 | } |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 288 | up_write(&mm->mmap_sem); |
| 289 | |
| 290 | mm_populate(ctx->mmap_base, populate); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 291 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 292 | pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base); |
| 293 | ctx->nr_pages = get_user_pages(current, mm, ctx->mmap_base, nr_pages, |
| 294 | 1, 0, ctx->ring_pages, NULL); |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 295 | for (i = 0; i < ctx->nr_pages; i++) |
| 296 | put_page(ctx->ring_pages[i]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 297 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 298 | if (unlikely(ctx->nr_pages != nr_pages)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 299 | aio_free_ring(ctx); |
| 300 | return -EAGAIN; |
| 301 | } |
| 302 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 303 | ctx->user_id = ctx->mmap_base; |
| 304 | ctx->nr_events = nr_events; /* trusted copy */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 305 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 306 | ring = kmap_atomic(ctx->ring_pages[0]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 307 | ring->nr = nr_events; /* user copy */ |
| 308 | ring->id = ctx->user_id; |
| 309 | ring->head = ring->tail = 0; |
| 310 | ring->magic = AIO_RING_MAGIC; |
| 311 | ring->compat_features = AIO_RING_COMPAT_FEATURES; |
| 312 | ring->incompat_features = AIO_RING_INCOMPAT_FEATURES; |
| 313 | ring->header_length = sizeof(struct aio_ring); |
Cong Wang | e8e3c3d | 2011-11-25 23:14:27 +0800 | [diff] [blame] | 314 | kunmap_atomic(ring); |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 315 | flush_dcache_page(ctx->ring_pages[0]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 316 | |
| 317 | return 0; |
| 318 | } |
| 319 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 320 | #define AIO_EVENTS_PER_PAGE (PAGE_SIZE / sizeof(struct io_event)) |
| 321 | #define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event)) |
| 322 | #define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE) |
| 323 | |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 324 | void kiocb_set_cancel_fn(struct kiocb *req, kiocb_cancel_fn *cancel) |
| 325 | { |
| 326 | struct kioctx *ctx = req->ki_ctx; |
| 327 | unsigned long flags; |
| 328 | |
| 329 | spin_lock_irqsave(&ctx->ctx_lock, flags); |
| 330 | |
| 331 | if (!req->ki_list.next) |
| 332 | list_add(&req->ki_list, &ctx->active_reqs); |
| 333 | |
| 334 | req->ki_cancel = cancel; |
| 335 | |
| 336 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); |
| 337 | } |
| 338 | EXPORT_SYMBOL(kiocb_set_cancel_fn); |
| 339 | |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 340 | static int kiocb_cancel(struct kioctx *ctx, struct kiocb *kiocb, |
| 341 | struct io_event *res) |
| 342 | { |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 343 | kiocb_cancel_fn *old, *cancel; |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 344 | int ret = -EINVAL; |
| 345 | |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 346 | /* |
| 347 | * Don't want to set kiocb->ki_cancel = KIOCB_CANCELLED unless it |
| 348 | * actually has a cancel function, hence the cmpxchg() |
| 349 | */ |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 350 | |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 351 | cancel = ACCESS_ONCE(kiocb->ki_cancel); |
| 352 | do { |
| 353 | if (!cancel || cancel == KIOCB_CANCELLED) |
| 354 | return ret; |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 355 | |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 356 | old = cancel; |
| 357 | cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED); |
| 358 | } while (cancel != old); |
| 359 | |
| 360 | atomic_inc(&kiocb->ki_users); |
| 361 | spin_unlock_irq(&ctx->ctx_lock); |
| 362 | |
| 363 | memset(res, 0, sizeof(*res)); |
| 364 | res->obj = (u64)(unsigned long)kiocb->ki_obj.user; |
| 365 | res->data = kiocb->ki_user_data; |
| 366 | ret = cancel(kiocb, res); |
| 367 | |
| 368 | spin_lock_irq(&ctx->ctx_lock); |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 369 | |
| 370 | return ret; |
| 371 | } |
| 372 | |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 373 | static void free_ioctx_rcu(struct rcu_head *head) |
| 374 | { |
| 375 | struct kioctx *ctx = container_of(head, struct kioctx, rcu_head); |
| 376 | kmem_cache_free(kioctx_cachep, ctx); |
| 377 | } |
| 378 | |
| 379 | /* |
| 380 | * When this function runs, the kioctx has been removed from the "hash table" |
| 381 | * and ctx->users has dropped to 0, so we know no more kiocbs can be submitted - |
| 382 | * now it's safe to cancel any that need to be. |
| 383 | */ |
| 384 | static void free_ioctx(struct kioctx *ctx) |
| 385 | { |
Kent Overstreet | 3e845ce | 2013-05-07 16:18:51 -0700 | [diff] [blame] | 386 | struct aio_ring *ring; |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 387 | struct io_event res; |
| 388 | struct kiocb *req; |
Kent Overstreet | 3e845ce | 2013-05-07 16:18:51 -0700 | [diff] [blame] | 389 | unsigned head, avail; |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 390 | |
| 391 | spin_lock_irq(&ctx->ctx_lock); |
| 392 | |
| 393 | while (!list_empty(&ctx->active_reqs)) { |
| 394 | req = list_first_entry(&ctx->active_reqs, |
| 395 | struct kiocb, ki_list); |
| 396 | |
| 397 | list_del_init(&req->ki_list); |
| 398 | kiocb_cancel(ctx, req, &res); |
| 399 | } |
| 400 | |
| 401 | spin_unlock_irq(&ctx->ctx_lock); |
| 402 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 403 | ring = kmap_atomic(ctx->ring_pages[0]); |
Kent Overstreet | 3e845ce | 2013-05-07 16:18:51 -0700 | [diff] [blame] | 404 | head = ring->head; |
| 405 | kunmap_atomic(ring); |
| 406 | |
| 407 | while (atomic_read(&ctx->reqs_active) > 0) { |
Benjamin LaHaise | 03e04f0 | 2013-05-24 15:55:38 -0700 | [diff] [blame] | 408 | wait_event(ctx->wait, |
| 409 | head != ctx->tail || |
| 410 | atomic_read(&ctx->reqs_active) <= 0); |
Kent Overstreet | 3e845ce | 2013-05-07 16:18:51 -0700 | [diff] [blame] | 411 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 412 | avail = (head <= ctx->tail ? ctx->tail : ctx->nr_events) - head; |
Kent Overstreet | 3e845ce | 2013-05-07 16:18:51 -0700 | [diff] [blame] | 413 | |
| 414 | atomic_sub(avail, &ctx->reqs_active); |
| 415 | head += avail; |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 416 | head %= ctx->nr_events; |
Kent Overstreet | 3e845ce | 2013-05-07 16:18:51 -0700 | [diff] [blame] | 417 | } |
| 418 | |
| 419 | WARN_ON(atomic_read(&ctx->reqs_active) < 0); |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 420 | |
| 421 | aio_free_ring(ctx); |
| 422 | |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 423 | pr_debug("freeing %p\n", ctx); |
| 424 | |
| 425 | /* |
| 426 | * Here the call_rcu() is between the wait_event() for reqs_active to |
| 427 | * hit 0, and freeing the ioctx. |
| 428 | * |
| 429 | * aio_complete() decrements reqs_active, but it has to touch the ioctx |
| 430 | * after to issue a wakeup so we use rcu. |
| 431 | */ |
| 432 | call_rcu(&ctx->rcu_head, free_ioctx_rcu); |
| 433 | } |
| 434 | |
| 435 | static void put_ioctx(struct kioctx *ctx) |
| 436 | { |
| 437 | if (unlikely(atomic_dec_and_test(&ctx->users))) |
| 438 | free_ioctx(ctx); |
| 439 | } |
| 440 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 441 | /* ioctx_alloc |
| 442 | * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed. |
| 443 | */ |
| 444 | static struct kioctx *ioctx_alloc(unsigned nr_events) |
| 445 | { |
Zach Brown | 41003a7 | 2013-05-07 16:18:25 -0700 | [diff] [blame] | 446 | struct mm_struct *mm = current->mm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 447 | struct kioctx *ctx; |
Al Viro | e23754f | 2012-03-06 14:33:22 -0500 | [diff] [blame] | 448 | int err = -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 449 | |
| 450 | /* Prevent overflows */ |
| 451 | if ((nr_events > (0x10000000U / sizeof(struct io_event))) || |
| 452 | (nr_events > (0x10000000U / sizeof(struct kiocb)))) { |
| 453 | pr_debug("ENOMEM: nr_events too high\n"); |
| 454 | return ERR_PTR(-EINVAL); |
| 455 | } |
| 456 | |
Al Viro | 2dd542b | 2012-03-10 23:10:35 -0500 | [diff] [blame] | 457 | if (!nr_events || (unsigned long)nr_events > aio_max_nr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 458 | return ERR_PTR(-EAGAIN); |
| 459 | |
Robert P. J. Day | c376222 | 2007-02-10 01:45:03 -0800 | [diff] [blame] | 460 | ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 461 | if (!ctx) |
| 462 | return ERR_PTR(-ENOMEM); |
| 463 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 464 | ctx->max_reqs = nr_events; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 465 | |
Al Viro | 86b62a2 | 2012-03-07 05:16:35 +0000 | [diff] [blame] | 466 | atomic_set(&ctx->users, 2); |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 467 | atomic_set(&ctx->dead, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 468 | spin_lock_init(&ctx->ctx_lock); |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 469 | spin_lock_init(&ctx->completion_lock); |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 470 | mutex_init(&ctx->ring_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 471 | init_waitqueue_head(&ctx->wait); |
| 472 | |
| 473 | INIT_LIST_HEAD(&ctx->active_reqs); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 474 | |
| 475 | if (aio_setup_ring(ctx) < 0) |
| 476 | goto out_freectx; |
| 477 | |
| 478 | /* limit the number of system wide aios */ |
Al Viro | 9fa1cb3 | 2012-03-10 23:14:05 -0500 | [diff] [blame] | 479 | spin_lock(&aio_nr_lock); |
Al Viro | 2dd542b | 2012-03-10 23:10:35 -0500 | [diff] [blame] | 480 | if (aio_nr + nr_events > aio_max_nr || |
| 481 | aio_nr + nr_events < aio_nr) { |
Al Viro | 9fa1cb3 | 2012-03-10 23:14:05 -0500 | [diff] [blame] | 482 | spin_unlock(&aio_nr_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 483 | goto out_cleanup; |
Al Viro | 2dd542b | 2012-03-10 23:10:35 -0500 | [diff] [blame] | 484 | } |
| 485 | aio_nr += ctx->max_reqs; |
Al Viro | 9fa1cb3 | 2012-03-10 23:14:05 -0500 | [diff] [blame] | 486 | spin_unlock(&aio_nr_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 487 | |
Jeff Moyer | 39fa003 | 2008-04-29 01:03:48 -0700 | [diff] [blame] | 488 | /* now link into global list. */ |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 489 | spin_lock(&mm->ioctx_lock); |
| 490 | hlist_add_head_rcu(&ctx->list, &mm->ioctx_list); |
| 491 | spin_unlock(&mm->ioctx_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 492 | |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 493 | pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n", |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 494 | ctx, ctx->user_id, mm, ctx->nr_events); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 495 | return ctx; |
| 496 | |
| 497 | out_cleanup: |
Al Viro | e23754f | 2012-03-06 14:33:22 -0500 | [diff] [blame] | 498 | err = -EAGAIN; |
| 499 | aio_free_ring(ctx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 500 | out_freectx: |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 501 | if (ctx->aio_ring_file) |
| 502 | fput(ctx->aio_ring_file); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 503 | kmem_cache_free(kioctx_cachep, ctx); |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 504 | pr_debug("error allocating ioctx %d\n", err); |
Al Viro | e23754f | 2012-03-06 14:33:22 -0500 | [diff] [blame] | 505 | return ERR_PTR(err); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 506 | } |
| 507 | |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 508 | static void kill_ioctx_work(struct work_struct *work) |
| 509 | { |
| 510 | struct kioctx *ctx = container_of(work, struct kioctx, rcu_work); |
| 511 | |
| 512 | wake_up_all(&ctx->wait); |
| 513 | put_ioctx(ctx); |
| 514 | } |
| 515 | |
| 516 | static void kill_ioctx_rcu(struct rcu_head *head) |
| 517 | { |
| 518 | struct kioctx *ctx = container_of(head, struct kioctx, rcu_head); |
| 519 | |
| 520 | INIT_WORK(&ctx->rcu_work, kill_ioctx_work); |
| 521 | schedule_work(&ctx->rcu_work); |
| 522 | } |
| 523 | |
| 524 | /* kill_ioctx |
| 525 | * Cancels all outstanding aio requests on an aio context. Used |
| 526 | * when the processes owning a context have all exited to encourage |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 527 | * the rapid destruction of the kioctx. |
| 528 | */ |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 529 | static void kill_ioctx(struct kioctx *ctx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 530 | { |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 531 | if (!atomic_xchg(&ctx->dead, 1)) { |
| 532 | hlist_del_rcu(&ctx->list); |
Al Viro | 06af121 | 2012-03-20 16:26:24 -0400 | [diff] [blame] | 533 | |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 534 | /* |
Kent Overstreet | 4fcc712 | 2013-06-12 14:04:59 -0700 | [diff] [blame] | 535 | * It'd be more correct to do this in free_ioctx(), after all |
| 536 | * the outstanding kiocbs have finished - but by then io_destroy |
| 537 | * has already returned, so io_setup() could potentially return |
| 538 | * -EAGAIN with no ioctxs actually in use (as far as userspace |
| 539 | * could tell). |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 540 | */ |
Kent Overstreet | 4fcc712 | 2013-06-12 14:04:59 -0700 | [diff] [blame] | 541 | spin_lock(&aio_nr_lock); |
| 542 | BUG_ON(aio_nr - ctx->max_reqs > aio_nr); |
| 543 | aio_nr -= ctx->max_reqs; |
| 544 | spin_unlock(&aio_nr_lock); |
| 545 | |
| 546 | if (ctx->mmap_size) |
| 547 | vm_munmap(ctx->mmap_base, ctx->mmap_size); |
| 548 | |
| 549 | /* Between hlist_del_rcu() and dropping the initial ref */ |
| 550 | call_rcu(&ctx->rcu_head, kill_ioctx_rcu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 551 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 552 | } |
| 553 | |
| 554 | /* wait_on_sync_kiocb: |
| 555 | * Waits on the given sync kiocb to complete. |
| 556 | */ |
Harvey Harrison | fc9b52c | 2008-02-08 04:19:52 -0800 | [diff] [blame] | 557 | ssize_t wait_on_sync_kiocb(struct kiocb *iocb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 558 | { |
Kent Overstreet | 11599eb | 2013-05-07 16:18:39 -0700 | [diff] [blame] | 559 | while (atomic_read(&iocb->ki_users)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 560 | set_current_state(TASK_UNINTERRUPTIBLE); |
Kent Overstreet | 11599eb | 2013-05-07 16:18:39 -0700 | [diff] [blame] | 561 | if (!atomic_read(&iocb->ki_users)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 562 | break; |
Jeff Moyer | 41d10da | 2007-10-16 23:27:20 -0700 | [diff] [blame] | 563 | io_schedule(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 564 | } |
| 565 | __set_current_state(TASK_RUNNING); |
| 566 | return iocb->ki_user_data; |
| 567 | } |
H Hartley Sweeten | 385773e | 2009-09-22 16:43:53 -0700 | [diff] [blame] | 568 | EXPORT_SYMBOL(wait_on_sync_kiocb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 569 | |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 570 | /* |
| 571 | * exit_aio: called when the last user of mm goes away. At this point, there is |
| 572 | * no way for any new requests to be submited or any of the io_* syscalls to be |
| 573 | * called on the context. |
| 574 | * |
| 575 | * There may be outstanding kiocbs, but free_ioctx() will explicitly wait on |
| 576 | * them. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 577 | */ |
Harvey Harrison | fc9b52c | 2008-02-08 04:19:52 -0800 | [diff] [blame] | 578 | void exit_aio(struct mm_struct *mm) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 579 | { |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 580 | struct kioctx *ctx; |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 581 | struct hlist_node *n; |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 582 | |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 583 | hlist_for_each_entry_safe(ctx, n, &mm->ioctx_list, list) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 584 | if (1 != atomic_read(&ctx->users)) |
| 585 | printk(KERN_DEBUG |
| 586 | "exit_aio:ioctx still alive: %d %d %d\n", |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 587 | atomic_read(&ctx->users), |
| 588 | atomic_read(&ctx->dead), |
Kent Overstreet | 11599eb | 2013-05-07 16:18:39 -0700 | [diff] [blame] | 589 | atomic_read(&ctx->reqs_active)); |
Al Viro | 936af15 | 2012-04-20 21:49:41 -0400 | [diff] [blame] | 590 | /* |
| 591 | * We don't need to bother with munmap() here - |
| 592 | * exit_mmap(mm) is coming and it'll unmap everything. |
| 593 | * Since aio_free_ring() uses non-zero ->mmap_size |
| 594 | * as indicator that it needs to unmap the area, |
| 595 | * just set it to 0; aio_free_ring() is the only |
| 596 | * place that uses ->mmap_size, so it's safe. |
Al Viro | 936af15 | 2012-04-20 21:49:41 -0400 | [diff] [blame] | 597 | */ |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 598 | ctx->mmap_size = 0; |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 599 | |
Kent Overstreet | 4fcc712 | 2013-06-12 14:04:59 -0700 | [diff] [blame] | 600 | kill_ioctx(ctx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 601 | } |
| 602 | } |
| 603 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 604 | /* aio_get_req |
Kent Overstreet | 11599eb | 2013-05-07 16:18:39 -0700 | [diff] [blame] | 605 | * Allocate a slot for an aio request. Increments the ki_users count |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 606 | * of the kioctx so that the kioctx stays around until all requests are |
| 607 | * complete. Returns NULL if no requests are free. |
| 608 | * |
Kent Overstreet | 11599eb | 2013-05-07 16:18:39 -0700 | [diff] [blame] | 609 | * Returns with kiocb->ki_users set to 2. The io submit code path holds |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 610 | * an extra reference while submitting the i/o. |
| 611 | * This prevents races between the aio code path referencing the |
| 612 | * req (after submitting it) and aio_complete() freeing the req. |
| 613 | */ |
Kent Overstreet | a1c8eae | 2013-05-07 16:18:53 -0700 | [diff] [blame] | 614 | static inline struct kiocb *aio_get_req(struct kioctx *ctx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 615 | { |
Kent Overstreet | a1c8eae | 2013-05-07 16:18:53 -0700 | [diff] [blame] | 616 | struct kiocb *req; |
| 617 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 618 | if (atomic_read(&ctx->reqs_active) >= ctx->nr_events) |
Kent Overstreet | a1c8eae | 2013-05-07 16:18:53 -0700 | [diff] [blame] | 619 | return NULL; |
| 620 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 621 | if (atomic_inc_return(&ctx->reqs_active) > ctx->nr_events - 1) |
Kent Overstreet | a1c8eae | 2013-05-07 16:18:53 -0700 | [diff] [blame] | 622 | goto out_put; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 623 | |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 624 | req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 625 | if (unlikely(!req)) |
Kent Overstreet | a1c8eae | 2013-05-07 16:18:53 -0700 | [diff] [blame] | 626 | goto out_put; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 627 | |
Kent Overstreet | 11599eb | 2013-05-07 16:18:39 -0700 | [diff] [blame] | 628 | atomic_set(&req->ki_users, 2); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 629 | req->ki_ctx = ctx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 630 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 631 | return req; |
Kent Overstreet | a1c8eae | 2013-05-07 16:18:53 -0700 | [diff] [blame] | 632 | out_put: |
| 633 | atomic_dec(&ctx->reqs_active); |
| 634 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 635 | } |
| 636 | |
Kent Overstreet | 11599eb | 2013-05-07 16:18:39 -0700 | [diff] [blame] | 637 | static void kiocb_free(struct kiocb *req) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 638 | { |
Kent Overstreet | 1d98ebf | 2013-05-07 16:18:37 -0700 | [diff] [blame] | 639 | if (req->ki_filp) |
| 640 | fput(req->ki_filp); |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 641 | if (req->ki_eventfd != NULL) |
| 642 | eventfd_ctx_put(req->ki_eventfd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 643 | if (req->ki_dtor) |
| 644 | req->ki_dtor(req); |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 645 | if (req->ki_iovec != &req->ki_inline_vec) |
| 646 | kfree(req->ki_iovec); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 647 | kmem_cache_free(kiocb_cachep, req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 648 | } |
| 649 | |
Kent Overstreet | 2d68449 | 2013-05-07 16:18:29 -0700 | [diff] [blame] | 650 | void aio_put_req(struct kiocb *req) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 651 | { |
Kent Overstreet | 11599eb | 2013-05-07 16:18:39 -0700 | [diff] [blame] | 652 | if (atomic_dec_and_test(&req->ki_users)) |
| 653 | kiocb_free(req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 654 | } |
H Hartley Sweeten | 385773e | 2009-09-22 16:43:53 -0700 | [diff] [blame] | 655 | EXPORT_SYMBOL(aio_put_req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 656 | |
Adrian Bunk | d5470b5 | 2008-04-29 00:58:57 -0700 | [diff] [blame] | 657 | static struct kioctx *lookup_ioctx(unsigned long ctx_id) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 658 | { |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 659 | struct mm_struct *mm = current->mm; |
Jeff Moyer | 65c2449 | 2009-03-18 17:04:21 -0700 | [diff] [blame] | 660 | struct kioctx *ctx, *ret = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 661 | |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 662 | rcu_read_lock(); |
| 663 | |
Sasha Levin | b67bfe0 | 2013-02-27 17:06:00 -0800 | [diff] [blame] | 664 | hlist_for_each_entry_rcu(ctx, &mm->ioctx_list, list) { |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 665 | if (ctx->user_id == ctx_id) { |
| 666 | atomic_inc(&ctx->users); |
Jeff Moyer | 65c2449 | 2009-03-18 17:04:21 -0700 | [diff] [blame] | 667 | ret = ctx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 668 | break; |
| 669 | } |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 670 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 671 | |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 672 | rcu_read_unlock(); |
Jeff Moyer | 65c2449 | 2009-03-18 17:04:21 -0700 | [diff] [blame] | 673 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 674 | } |
| 675 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 676 | /* aio_complete |
| 677 | * Called when the io request on the given iocb is complete. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 678 | */ |
Kent Overstreet | 2d68449 | 2013-05-07 16:18:29 -0700 | [diff] [blame] | 679 | void aio_complete(struct kiocb *iocb, long res, long res2) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 680 | { |
| 681 | struct kioctx *ctx = iocb->ki_ctx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 682 | struct aio_ring *ring; |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 683 | struct io_event *ev_page, *event; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 684 | unsigned long flags; |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 685 | unsigned tail, pos; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 686 | |
Zach Brown | 20dcae3 | 2005-11-13 16:07:33 -0800 | [diff] [blame] | 687 | /* |
| 688 | * Special case handling for sync iocbs: |
| 689 | * - events go directly into the iocb for fast handling |
| 690 | * - the sync task with the iocb in its stack holds the single iocb |
| 691 | * ref, no other paths have a way to get another ref |
| 692 | * - the sync task helpfully left a reference to itself in the iocb |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 693 | */ |
| 694 | if (is_sync_kiocb(iocb)) { |
Kent Overstreet | 11599eb | 2013-05-07 16:18:39 -0700 | [diff] [blame] | 695 | BUG_ON(atomic_read(&iocb->ki_users) != 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 696 | iocb->ki_user_data = res; |
Kent Overstreet | 11599eb | 2013-05-07 16:18:39 -0700 | [diff] [blame] | 697 | atomic_set(&iocb->ki_users, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 698 | wake_up_process(iocb->ki_obj.tsk); |
Kent Overstreet | 2d68449 | 2013-05-07 16:18:29 -0700 | [diff] [blame] | 699 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 700 | } |
| 701 | |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 702 | /* |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 703 | * Take rcu_read_lock() in case the kioctx is being destroyed, as we |
| 704 | * need to issue a wakeup after decrementing reqs_active. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 705 | */ |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 706 | rcu_read_lock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 707 | |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 708 | if (iocb->ki_list.next) { |
| 709 | unsigned long flags; |
| 710 | |
| 711 | spin_lock_irqsave(&ctx->ctx_lock, flags); |
| 712 | list_del(&iocb->ki_list); |
| 713 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); |
| 714 | } |
Kent Overstreet | 11599eb | 2013-05-07 16:18:39 -0700 | [diff] [blame] | 715 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 716 | /* |
| 717 | * cancelled requests don't get events, userland was given one |
| 718 | * when the event got cancelled. |
| 719 | */ |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 720 | if (unlikely(xchg(&iocb->ki_cancel, |
Kent Overstreet | 3e845ce | 2013-05-07 16:18:51 -0700 | [diff] [blame] | 721 | KIOCB_CANCELLED) == KIOCB_CANCELLED)) { |
| 722 | atomic_dec(&ctx->reqs_active); |
| 723 | /* Still need the wake_up in case free_ioctx is waiting */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 724 | goto put_rq; |
Kent Overstreet | 3e845ce | 2013-05-07 16:18:51 -0700 | [diff] [blame] | 725 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 726 | |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 727 | /* |
| 728 | * Add a completion event to the ring buffer. Must be done holding |
Tang Chen | 4b30f07 | 2013-07-03 15:09:16 -0700 | [diff] [blame] | 729 | * ctx->completion_lock to prevent other code from messing with the tail |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 730 | * pointer since we might be called from irq context. |
| 731 | */ |
| 732 | spin_lock_irqsave(&ctx->completion_lock, flags); |
| 733 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 734 | tail = ctx->tail; |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 735 | pos = tail + AIO_EVENTS_OFFSET; |
| 736 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 737 | if (++tail >= ctx->nr_events) |
Ken Chen | 4bf69b2 | 2005-05-01 08:59:15 -0700 | [diff] [blame] | 738 | tail = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 739 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 740 | ev_page = kmap_atomic(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 741 | event = ev_page + pos % AIO_EVENTS_PER_PAGE; |
| 742 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 743 | event->obj = (u64)(unsigned long)iocb->ki_obj.user; |
| 744 | event->data = iocb->ki_user_data; |
| 745 | event->res = res; |
| 746 | event->res2 = res2; |
| 747 | |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 748 | kunmap_atomic(ev_page); |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 749 | flush_dcache_page(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 750 | |
| 751 | pr_debug("%p[%u]: %p: %p %Lx %lx %lx\n", |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 752 | ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data, |
| 753 | res, res2); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 754 | |
| 755 | /* after flagging the request as done, we |
| 756 | * must never even look at it again |
| 757 | */ |
| 758 | smp_wmb(); /* make event visible before updating tail */ |
| 759 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 760 | ctx->tail = tail; |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 761 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 762 | ring = kmap_atomic(ctx->ring_pages[0]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 763 | ring->tail = tail; |
Cong Wang | e8e3c3d | 2011-11-25 23:14:27 +0800 | [diff] [blame] | 764 | kunmap_atomic(ring); |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 765 | flush_dcache_page(ctx->ring_pages[0]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 766 | |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 767 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
| 768 | |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 769 | pr_debug("added to ring %p at [%u]\n", iocb, tail); |
Davide Libenzi | 8d1c98b | 2008-04-10 21:29:19 -0700 | [diff] [blame] | 770 | |
| 771 | /* |
| 772 | * Check if the user asked us to deliver the result through an |
| 773 | * eventfd. The eventfd_signal() function is safe to be called |
| 774 | * from IRQ context. |
| 775 | */ |
Davide Libenzi | 87c3a86 | 2009-03-18 17:04:19 -0700 | [diff] [blame] | 776 | if (iocb->ki_eventfd != NULL) |
Davide Libenzi | 8d1c98b | 2008-04-10 21:29:19 -0700 | [diff] [blame] | 777 | eventfd_signal(iocb->ki_eventfd, 1); |
| 778 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 779 | put_rq: |
| 780 | /* everything turned out well, dispose of the aiocb. */ |
Kent Overstreet | 11599eb | 2013-05-07 16:18:39 -0700 | [diff] [blame] | 781 | aio_put_req(iocb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 782 | |
Quentin Barnes | 6cb2a21 | 2008-03-19 17:00:39 -0700 | [diff] [blame] | 783 | /* |
| 784 | * We have to order our ring_info tail store above and test |
| 785 | * of the wait list below outside the wait lock. This is |
| 786 | * like in wake_up_bit() where clearing a bit has to be |
| 787 | * ordered with the unlocked test. |
| 788 | */ |
| 789 | smp_mb(); |
| 790 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 791 | if (waitqueue_active(&ctx->wait)) |
| 792 | wake_up(&ctx->wait); |
| 793 | |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 794 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 795 | } |
H Hartley Sweeten | 385773e | 2009-09-22 16:43:53 -0700 | [diff] [blame] | 796 | EXPORT_SYMBOL(aio_complete); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 797 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 798 | /* aio_read_events |
| 799 | * Pull an event off of the ioctx's event ring. Returns the number of |
| 800 | * events fetched |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 801 | */ |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 802 | static long aio_read_events_ring(struct kioctx *ctx, |
| 803 | struct io_event __user *event, long nr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 804 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 805 | struct aio_ring *ring; |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 806 | unsigned head, pos; |
| 807 | long ret = 0; |
| 808 | int copy_ret; |
| 809 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 810 | mutex_lock(&ctx->ring_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 811 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 812 | ring = kmap_atomic(ctx->ring_pages[0]); |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 813 | head = ring->head; |
| 814 | kunmap_atomic(ring); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 815 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 816 | pr_debug("h%u t%u m%u\n", head, ctx->tail, ctx->nr_events); |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 817 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 818 | if (head == ctx->tail) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 819 | goto out; |
| 820 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 821 | while (ret < nr) { |
| 822 | long avail; |
| 823 | struct io_event *ev; |
| 824 | struct page *page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 825 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 826 | avail = (head <= ctx->tail ? ctx->tail : ctx->nr_events) - head; |
| 827 | if (head == ctx->tail) |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 828 | break; |
| 829 | |
| 830 | avail = min(avail, nr - ret); |
| 831 | avail = min_t(long, avail, AIO_EVENTS_PER_PAGE - |
| 832 | ((head + AIO_EVENTS_OFFSET) % AIO_EVENTS_PER_PAGE)); |
| 833 | |
| 834 | pos = head + AIO_EVENTS_OFFSET; |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 835 | page = ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]; |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 836 | pos %= AIO_EVENTS_PER_PAGE; |
| 837 | |
| 838 | ev = kmap(page); |
| 839 | copy_ret = copy_to_user(event + ret, ev + pos, |
| 840 | sizeof(*ev) * avail); |
| 841 | kunmap(page); |
| 842 | |
| 843 | if (unlikely(copy_ret)) { |
| 844 | ret = -EFAULT; |
| 845 | goto out; |
| 846 | } |
| 847 | |
| 848 | ret += avail; |
| 849 | head += avail; |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 850 | head %= ctx->nr_events; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 851 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 852 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 853 | ring = kmap_atomic(ctx->ring_pages[0]); |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 854 | ring->head = head; |
Zhao Hongjiang | 91d80a8 | 2013-04-26 11:03:53 +0800 | [diff] [blame] | 855 | kunmap_atomic(ring); |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 856 | flush_dcache_page(ctx->ring_pages[0]); |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 857 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 858 | pr_debug("%li h%u t%u\n", ret, head, ctx->tail); |
Kent Overstreet | 3e845ce | 2013-05-07 16:18:51 -0700 | [diff] [blame] | 859 | |
| 860 | atomic_sub(ret, &ctx->reqs_active); |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 861 | out: |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 862 | mutex_unlock(&ctx->ring_lock); |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 863 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 864 | return ret; |
| 865 | } |
| 866 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 867 | static bool aio_read_events(struct kioctx *ctx, long min_nr, long nr, |
| 868 | struct io_event __user *event, long *i) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 869 | { |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 870 | long ret = aio_read_events_ring(ctx, event + *i, nr - *i); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 871 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 872 | if (ret > 0) |
| 873 | *i += ret; |
| 874 | |
| 875 | if (unlikely(atomic_read(&ctx->dead))) |
| 876 | ret = -EINVAL; |
| 877 | |
| 878 | if (!*i) |
| 879 | *i = ret; |
| 880 | |
| 881 | return ret < 0 || *i >= min_nr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 882 | } |
| 883 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 884 | static long read_events(struct kioctx *ctx, long min_nr, long nr, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 885 | struct io_event __user *event, |
| 886 | struct timespec __user *timeout) |
| 887 | { |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 888 | ktime_t until = { .tv64 = KTIME_MAX }; |
| 889 | long ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 890 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 891 | if (timeout) { |
| 892 | struct timespec ts; |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 893 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 894 | if (unlikely(copy_from_user(&ts, timeout, sizeof(ts)))) |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 895 | return -EFAULT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 896 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 897 | until = timespec_to_ktime(ts); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 898 | } |
| 899 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 900 | /* |
| 901 | * Note that aio_read_events() is being called as the conditional - i.e. |
| 902 | * we're calling it after prepare_to_wait() has set task state to |
| 903 | * TASK_INTERRUPTIBLE. |
| 904 | * |
| 905 | * But aio_read_events() can block, and if it blocks it's going to flip |
| 906 | * the task state back to TASK_RUNNING. |
| 907 | * |
| 908 | * This should be ok, provided it doesn't flip the state back to |
| 909 | * TASK_RUNNING and return 0 too much - that causes us to spin. That |
| 910 | * will only happen if the mutex_lock() call blocks, and we then find |
| 911 | * the ringbuffer empty. So in practice we should be ok, but it's |
| 912 | * something to be aware of when touching this code. |
| 913 | */ |
| 914 | wait_event_interruptible_hrtimeout(ctx->wait, |
| 915 | aio_read_events(ctx, min_nr, nr, event, &ret), until); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 916 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 917 | if (!ret && signal_pending(current)) |
| 918 | ret = -EINTR; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 919 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 920 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 921 | } |
| 922 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 923 | /* sys_io_setup: |
| 924 | * Create an aio_context capable of receiving at least nr_events. |
| 925 | * ctxp must not point to an aio_context that already exists, and |
| 926 | * must be initialized to 0 prior to the call. On successful |
| 927 | * creation of the aio_context, *ctxp is filled in with the resulting |
| 928 | * handle. May fail with -EINVAL if *ctxp is not initialized, |
| 929 | * if the specified nr_events exceeds internal limits. May fail |
| 930 | * with -EAGAIN if the specified nr_events exceeds the user's limit |
| 931 | * of available events. May fail with -ENOMEM if insufficient kernel |
| 932 | * resources are available. May fail with -EFAULT if an invalid |
| 933 | * pointer is passed for ctxp. Will fail with -ENOSYS if not |
| 934 | * implemented. |
| 935 | */ |
Heiko Carstens | 002c897 | 2009-01-14 14:14:18 +0100 | [diff] [blame] | 936 | SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 937 | { |
| 938 | struct kioctx *ioctx = NULL; |
| 939 | unsigned long ctx; |
| 940 | long ret; |
| 941 | |
| 942 | ret = get_user(ctx, ctxp); |
| 943 | if (unlikely(ret)) |
| 944 | goto out; |
| 945 | |
| 946 | ret = -EINVAL; |
Zach Brown | d55b5fd | 2005-11-07 00:59:31 -0800 | [diff] [blame] | 947 | if (unlikely(ctx || nr_events == 0)) { |
| 948 | pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n", |
| 949 | ctx, nr_events); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 950 | goto out; |
| 951 | } |
| 952 | |
| 953 | ioctx = ioctx_alloc(nr_events); |
| 954 | ret = PTR_ERR(ioctx); |
| 955 | if (!IS_ERR(ioctx)) { |
| 956 | ret = put_user(ioctx->user_id, ctxp); |
Al Viro | a2e1859 | 2012-03-20 16:27:57 -0400 | [diff] [blame] | 957 | if (ret) |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 958 | kill_ioctx(ioctx); |
Al Viro | a2e1859 | 2012-03-20 16:27:57 -0400 | [diff] [blame] | 959 | put_ioctx(ioctx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 960 | } |
| 961 | |
| 962 | out: |
| 963 | return ret; |
| 964 | } |
| 965 | |
| 966 | /* sys_io_destroy: |
| 967 | * Destroy the aio_context specified. May cancel any outstanding |
| 968 | * AIOs and block on completion. Will fail with -ENOSYS if not |
Satoru Takeuchi | 642b512 | 2010-08-05 11:23:11 -0700 | [diff] [blame] | 969 | * implemented. May fail with -EINVAL if the context pointed to |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 970 | * is invalid. |
| 971 | */ |
Heiko Carstens | 002c897 | 2009-01-14 14:14:18 +0100 | [diff] [blame] | 972 | SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 973 | { |
| 974 | struct kioctx *ioctx = lookup_ioctx(ctx); |
| 975 | if (likely(NULL != ioctx)) { |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 976 | kill_ioctx(ioctx); |
Al Viro | a2e1859 | 2012-03-20 16:27:57 -0400 | [diff] [blame] | 977 | put_ioctx(ioctx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 978 | return 0; |
| 979 | } |
| 980 | pr_debug("EINVAL: io_destroy: invalid context id\n"); |
| 981 | return -EINVAL; |
| 982 | } |
| 983 | |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 984 | static void aio_advance_iovec(struct kiocb *iocb, ssize_t ret) |
| 985 | { |
| 986 | struct iovec *iov = &iocb->ki_iovec[iocb->ki_cur_seg]; |
| 987 | |
| 988 | BUG_ON(ret <= 0); |
| 989 | |
| 990 | while (iocb->ki_cur_seg < iocb->ki_nr_segs && ret > 0) { |
| 991 | ssize_t this = min((ssize_t)iov->iov_len, ret); |
| 992 | iov->iov_base += this; |
| 993 | iov->iov_len -= this; |
| 994 | iocb->ki_left -= this; |
| 995 | ret -= this; |
| 996 | if (iov->iov_len == 0) { |
| 997 | iocb->ki_cur_seg++; |
| 998 | iov++; |
| 999 | } |
| 1000 | } |
| 1001 | |
| 1002 | /* the caller should not have done more io than what fit in |
| 1003 | * the remaining iovecs */ |
| 1004 | BUG_ON(ret > 0 && iocb->ki_left == 0); |
| 1005 | } |
| 1006 | |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1007 | typedef ssize_t (aio_rw_op)(struct kiocb *, const struct iovec *, |
| 1008 | unsigned long, loff_t); |
| 1009 | |
| 1010 | static ssize_t aio_rw_vect_retry(struct kiocb *iocb, int rw, aio_rw_op *rw_op) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1011 | { |
| 1012 | struct file *file = iocb->ki_filp; |
| 1013 | struct address_space *mapping = file->f_mapping; |
| 1014 | struct inode *inode = mapping->host; |
| 1015 | ssize_t ret = 0; |
| 1016 | |
Rusty Russell | c2ec668 | 2008-02-08 04:20:15 -0800 | [diff] [blame] | 1017 | /* This matches the pread()/pwrite() logic */ |
| 1018 | if (iocb->ki_pos < 0) |
| 1019 | return -EINVAL; |
| 1020 | |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1021 | if (rw == WRITE) |
Al Viro | 8d71db4 | 2013-03-19 21:01:03 -0400 | [diff] [blame] | 1022 | file_start_write(file); |
Zach Brown | 897f15f | 2005-09-30 11:58:55 -0700 | [diff] [blame] | 1023 | do { |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1024 | ret = rw_op(iocb, &iocb->ki_iovec[iocb->ki_cur_seg], |
| 1025 | iocb->ki_nr_segs - iocb->ki_cur_seg, |
| 1026 | iocb->ki_pos); |
| 1027 | if (ret > 0) |
| 1028 | aio_advance_iovec(iocb, ret); |
Badari Pulavarty | 027445c | 2006-09-30 23:28:46 -0700 | [diff] [blame] | 1029 | |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1030 | /* retry all partial writes. retry partial reads as long as its a |
| 1031 | * regular file. */ |
Zach Brown | 353fb07 | 2005-09-30 11:58:56 -0700 | [diff] [blame] | 1032 | } while (ret > 0 && iocb->ki_left > 0 && |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1033 | (rw == WRITE || |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1034 | (!S_ISFIFO(inode->i_mode) && !S_ISSOCK(inode->i_mode)))); |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1035 | if (rw == WRITE) |
Al Viro | 8d71db4 | 2013-03-19 21:01:03 -0400 | [diff] [blame] | 1036 | file_end_write(file); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1037 | |
| 1038 | /* This means we must have transferred all that we could */ |
| 1039 | /* No need to retry anymore */ |
| 1040 | if ((ret == 0) || (iocb->ki_left == 0)) |
| 1041 | ret = iocb->ki_nbytes - iocb->ki_left; |
| 1042 | |
Rusty Russell | 7adfa2f | 2008-02-08 04:20:14 -0800 | [diff] [blame] | 1043 | /* If we managed to write some out we return that, rather than |
| 1044 | * the eventual error. */ |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1045 | if (rw == WRITE |
Zach Brown | 41003a7 | 2013-05-07 16:18:25 -0700 | [diff] [blame] | 1046 | && ret < 0 && ret != -EIOCBQUEUED |
Rusty Russell | 7adfa2f | 2008-02-08 04:20:14 -0800 | [diff] [blame] | 1047 | && iocb->ki_nbytes - iocb->ki_left) |
| 1048 | ret = iocb->ki_nbytes - iocb->ki_left; |
| 1049 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1050 | return ret; |
| 1051 | } |
| 1052 | |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1053 | static ssize_t aio_setup_vectored_rw(int rw, struct kiocb *kiocb, bool compat) |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1054 | { |
| 1055 | ssize_t ret; |
| 1056 | |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1057 | kiocb->ki_nr_segs = kiocb->ki_nbytes; |
| 1058 | |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 1059 | #ifdef CONFIG_COMPAT |
| 1060 | if (compat) |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1061 | ret = compat_rw_copy_check_uvector(rw, |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 1062 | (struct compat_iovec __user *)kiocb->ki_buf, |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1063 | kiocb->ki_nr_segs, 1, &kiocb->ki_inline_vec, |
Christopher Yeoh | ac34ebb | 2012-05-31 16:26:42 -0700 | [diff] [blame] | 1064 | &kiocb->ki_iovec); |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 1065 | else |
| 1066 | #endif |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1067 | ret = rw_copy_check_uvector(rw, |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 1068 | (struct iovec __user *)kiocb->ki_buf, |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1069 | kiocb->ki_nr_segs, 1, &kiocb->ki_inline_vec, |
Christopher Yeoh | ac34ebb | 2012-05-31 16:26:42 -0700 | [diff] [blame] | 1070 | &kiocb->ki_iovec); |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1071 | if (ret < 0) |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1072 | return ret; |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1073 | |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1074 | /* ki_nbytes now reflect bytes instead of segs */ |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1075 | kiocb->ki_nbytes = ret; |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1076 | return 0; |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1077 | } |
| 1078 | |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1079 | static ssize_t aio_setup_single_vector(int rw, struct kiocb *kiocb) |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1080 | { |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1081 | if (unlikely(!access_ok(!rw, kiocb->ki_buf, kiocb->ki_nbytes))) |
| 1082 | return -EFAULT; |
Linus Torvalds | a70b52e | 2012-05-21 16:06:20 -0700 | [diff] [blame] | 1083 | |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1084 | kiocb->ki_iovec = &kiocb->ki_inline_vec; |
| 1085 | kiocb->ki_iovec->iov_base = kiocb->ki_buf; |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1086 | kiocb->ki_iovec->iov_len = kiocb->ki_nbytes; |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1087 | kiocb->ki_nr_segs = 1; |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1088 | return 0; |
| 1089 | } |
| 1090 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1091 | /* |
| 1092 | * aio_setup_iocb: |
| 1093 | * Performs the initial checks and aio retry method |
| 1094 | * setup for the kiocb at the time of io submission. |
| 1095 | */ |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1096 | static ssize_t aio_run_iocb(struct kiocb *req, bool compat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1097 | { |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1098 | struct file *file = req->ki_filp; |
| 1099 | ssize_t ret; |
| 1100 | int rw; |
| 1101 | fmode_t mode; |
| 1102 | aio_rw_op *rw_op; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1103 | |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1104 | switch (req->ki_opcode) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1105 | case IOCB_CMD_PREAD: |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1106 | case IOCB_CMD_PREADV: |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1107 | mode = FMODE_READ; |
| 1108 | rw = READ; |
| 1109 | rw_op = file->f_op->aio_read; |
| 1110 | goto rw_common; |
| 1111 | |
| 1112 | case IOCB_CMD_PWRITE: |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1113 | case IOCB_CMD_PWRITEV: |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1114 | mode = FMODE_WRITE; |
| 1115 | rw = WRITE; |
| 1116 | rw_op = file->f_op->aio_write; |
| 1117 | goto rw_common; |
| 1118 | rw_common: |
| 1119 | if (unlikely(!(file->f_mode & mode))) |
| 1120 | return -EBADF; |
| 1121 | |
| 1122 | if (!rw_op) |
| 1123 | return -EINVAL; |
| 1124 | |
| 1125 | ret = (req->ki_opcode == IOCB_CMD_PREADV || |
| 1126 | req->ki_opcode == IOCB_CMD_PWRITEV) |
| 1127 | ? aio_setup_vectored_rw(rw, req, compat) |
| 1128 | : aio_setup_single_vector(rw, req); |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1129 | if (ret) |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1130 | return ret; |
| 1131 | |
| 1132 | ret = rw_verify_area(rw, file, &req->ki_pos, req->ki_nbytes); |
| 1133 | if (ret < 0) |
| 1134 | return ret; |
| 1135 | |
| 1136 | req->ki_nbytes = ret; |
| 1137 | req->ki_left = ret; |
| 1138 | |
| 1139 | ret = aio_rw_vect_retry(req, rw, rw_op); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1140 | break; |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1141 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1142 | case IOCB_CMD_FDSYNC: |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1143 | if (!file->f_op->aio_fsync) |
| 1144 | return -EINVAL; |
| 1145 | |
| 1146 | ret = file->f_op->aio_fsync(req, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1147 | break; |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1148 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1149 | case IOCB_CMD_FSYNC: |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1150 | if (!file->f_op->aio_fsync) |
| 1151 | return -EINVAL; |
| 1152 | |
| 1153 | ret = file->f_op->aio_fsync(req, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1154 | break; |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1155 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1156 | default: |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 1157 | pr_debug("EINVAL: no operation provided\n"); |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1158 | return -EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1159 | } |
| 1160 | |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1161 | if (ret != -EIOCBQUEUED) { |
| 1162 | /* |
| 1163 | * There's no easy way to restart the syscall since other AIO's |
| 1164 | * may be already running. Just fail this IO with EINTR. |
| 1165 | */ |
| 1166 | if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR || |
| 1167 | ret == -ERESTARTNOHAND || |
| 1168 | ret == -ERESTART_RESTARTBLOCK)) |
| 1169 | ret = -EINTR; |
| 1170 | aio_complete(req, ret, 0); |
| 1171 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1172 | |
| 1173 | return 0; |
| 1174 | } |
| 1175 | |
Adrian Bunk | d5470b5 | 2008-04-29 00:58:57 -0700 | [diff] [blame] | 1176 | static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb, |
Kent Overstreet | a1c8eae | 2013-05-07 16:18:53 -0700 | [diff] [blame] | 1177 | struct iocb *iocb, bool compat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1178 | { |
| 1179 | struct kiocb *req; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1180 | ssize_t ret; |
| 1181 | |
| 1182 | /* enforce forwards compatibility on users */ |
Davide Libenzi | 9c3060b | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 1183 | if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) { |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 1184 | pr_debug("EINVAL: reserve field set\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1185 | return -EINVAL; |
| 1186 | } |
| 1187 | |
| 1188 | /* prevent overflows */ |
| 1189 | if (unlikely( |
| 1190 | (iocb->aio_buf != (unsigned long)iocb->aio_buf) || |
| 1191 | (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) || |
| 1192 | ((ssize_t)iocb->aio_nbytes < 0) |
| 1193 | )) { |
| 1194 | pr_debug("EINVAL: io_submit: overflow check\n"); |
| 1195 | return -EINVAL; |
| 1196 | } |
| 1197 | |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1198 | req = aio_get_req(ctx); |
Kent Overstreet | 1d98ebf | 2013-05-07 16:18:37 -0700 | [diff] [blame] | 1199 | if (unlikely(!req)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1200 | return -EAGAIN; |
Kent Overstreet | 1d98ebf | 2013-05-07 16:18:37 -0700 | [diff] [blame] | 1201 | |
| 1202 | req->ki_filp = fget(iocb->aio_fildes); |
| 1203 | if (unlikely(!req->ki_filp)) { |
| 1204 | ret = -EBADF; |
| 1205 | goto out_put_req; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1206 | } |
Kent Overstreet | 1d98ebf | 2013-05-07 16:18:37 -0700 | [diff] [blame] | 1207 | |
Davide Libenzi | 9c3060b | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 1208 | if (iocb->aio_flags & IOCB_FLAG_RESFD) { |
| 1209 | /* |
| 1210 | * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an |
| 1211 | * instance of the file* now. The file descriptor must be |
| 1212 | * an eventfd() fd, and will be signaled for each completed |
| 1213 | * event using the eventfd_signal() function. |
| 1214 | */ |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 1215 | req->ki_eventfd = eventfd_ctx_fdget((int) iocb->aio_resfd); |
Hirofumi Nakagawa | 801678c | 2008-04-29 01:03:09 -0700 | [diff] [blame] | 1216 | if (IS_ERR(req->ki_eventfd)) { |
Davide Libenzi | 9c3060b | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 1217 | ret = PTR_ERR(req->ki_eventfd); |
Davide Libenzi | 87c3a86 | 2009-03-18 17:04:19 -0700 | [diff] [blame] | 1218 | req->ki_eventfd = NULL; |
Davide Libenzi | 9c3060b | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 1219 | goto out_put_req; |
| 1220 | } |
| 1221 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1222 | |
Kent Overstreet | 8a66089 | 2013-05-07 16:19:10 -0700 | [diff] [blame] | 1223 | ret = put_user(KIOCB_KEY, &user_iocb->aio_key); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1224 | if (unlikely(ret)) { |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 1225 | pr_debug("EFAULT: aio_key\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1226 | goto out_put_req; |
| 1227 | } |
| 1228 | |
| 1229 | req->ki_obj.user = user_iocb; |
| 1230 | req->ki_user_data = iocb->aio_data; |
| 1231 | req->ki_pos = iocb->aio_offset; |
| 1232 | |
| 1233 | req->ki_buf = (char __user *)(unsigned long)iocb->aio_buf; |
| 1234 | req->ki_left = req->ki_nbytes = iocb->aio_nbytes; |
| 1235 | req->ki_opcode = iocb->aio_lio_opcode; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1236 | |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1237 | ret = aio_run_iocb(req, compat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1238 | if (ret) |
| 1239 | goto out_put_req; |
| 1240 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1241 | aio_put_req(req); /* drop extra ref to req */ |
| 1242 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1243 | out_put_req: |
Kent Overstreet | 11599eb | 2013-05-07 16:18:39 -0700 | [diff] [blame] | 1244 | atomic_dec(&ctx->reqs_active); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1245 | aio_put_req(req); /* drop extra ref to req */ |
| 1246 | aio_put_req(req); /* drop i/o ref to req */ |
| 1247 | return ret; |
| 1248 | } |
| 1249 | |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 1250 | long do_io_submit(aio_context_t ctx_id, long nr, |
| 1251 | struct iocb __user *__user *iocbpp, bool compat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1252 | { |
| 1253 | struct kioctx *ctx; |
| 1254 | long ret = 0; |
Jeff Moyer | 080d676 | 2011-11-02 13:40:10 -0700 | [diff] [blame] | 1255 | int i = 0; |
Shaohua Li | 9f5b942 | 2010-07-01 07:55:01 +0200 | [diff] [blame] | 1256 | struct blk_plug plug; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1257 | |
| 1258 | if (unlikely(nr < 0)) |
| 1259 | return -EINVAL; |
| 1260 | |
Jeff Moyer | 75e1c70 | 2010-09-10 14:16:00 -0700 | [diff] [blame] | 1261 | if (unlikely(nr > LONG_MAX/sizeof(*iocbpp))) |
| 1262 | nr = LONG_MAX/sizeof(*iocbpp); |
| 1263 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1264 | if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp))))) |
| 1265 | return -EFAULT; |
| 1266 | |
| 1267 | ctx = lookup_ioctx(ctx_id); |
| 1268 | if (unlikely(!ctx)) { |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 1269 | pr_debug("EINVAL: invalid context id\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1270 | return -EINVAL; |
| 1271 | } |
| 1272 | |
Shaohua Li | 9f5b942 | 2010-07-01 07:55:01 +0200 | [diff] [blame] | 1273 | blk_start_plug(&plug); |
| 1274 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1275 | /* |
| 1276 | * AKPM: should this return a partial result if some of the IOs were |
| 1277 | * successfully submitted? |
| 1278 | */ |
| 1279 | for (i=0; i<nr; i++) { |
| 1280 | struct iocb __user *user_iocb; |
| 1281 | struct iocb tmp; |
| 1282 | |
| 1283 | if (unlikely(__get_user(user_iocb, iocbpp + i))) { |
| 1284 | ret = -EFAULT; |
| 1285 | break; |
| 1286 | } |
| 1287 | |
| 1288 | if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) { |
| 1289 | ret = -EFAULT; |
| 1290 | break; |
| 1291 | } |
| 1292 | |
Kent Overstreet | a1c8eae | 2013-05-07 16:18:53 -0700 | [diff] [blame] | 1293 | ret = io_submit_one(ctx, user_iocb, &tmp, compat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1294 | if (ret) |
| 1295 | break; |
| 1296 | } |
Shaohua Li | 9f5b942 | 2010-07-01 07:55:01 +0200 | [diff] [blame] | 1297 | blk_finish_plug(&plug); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1298 | |
| 1299 | put_ioctx(ctx); |
| 1300 | return i ? i : ret; |
| 1301 | } |
| 1302 | |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 1303 | /* sys_io_submit: |
| 1304 | * Queue the nr iocbs pointed to by iocbpp for processing. Returns |
| 1305 | * the number of iocbs queued. May return -EINVAL if the aio_context |
| 1306 | * specified by ctx_id is invalid, if nr is < 0, if the iocb at |
| 1307 | * *iocbpp[0] is not properly initialized, if the operation specified |
| 1308 | * is invalid for the file descriptor in the iocb. May fail with |
| 1309 | * -EFAULT if any of the data structures point to invalid data. May |
| 1310 | * fail with -EBADF if the file descriptor specified in the first |
| 1311 | * iocb is invalid. May fail with -EAGAIN if insufficient resources |
| 1312 | * are available to queue any iocbs. Will return 0 if nr is 0. Will |
| 1313 | * fail with -ENOSYS if not implemented. |
| 1314 | */ |
| 1315 | SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr, |
| 1316 | struct iocb __user * __user *, iocbpp) |
| 1317 | { |
| 1318 | return do_io_submit(ctx_id, nr, iocbpp, 0); |
| 1319 | } |
| 1320 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1321 | /* lookup_kiocb |
| 1322 | * Finds a given iocb for cancellation. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1323 | */ |
Adrian Bunk | 25ee7e3 | 2005-04-25 08:18:14 -0700 | [diff] [blame] | 1324 | static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb, |
| 1325 | u32 key) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1326 | { |
| 1327 | struct list_head *pos; |
Zach Brown | d00689a | 2005-11-13 16:07:34 -0800 | [diff] [blame] | 1328 | |
| 1329 | assert_spin_locked(&ctx->ctx_lock); |
| 1330 | |
Kent Overstreet | 8a66089 | 2013-05-07 16:19:10 -0700 | [diff] [blame] | 1331 | if (key != KIOCB_KEY) |
| 1332 | return NULL; |
| 1333 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1334 | /* TODO: use a hash or array, this sucks. */ |
| 1335 | list_for_each(pos, &ctx->active_reqs) { |
| 1336 | struct kiocb *kiocb = list_kiocb(pos); |
Kent Overstreet | 8a66089 | 2013-05-07 16:19:10 -0700 | [diff] [blame] | 1337 | if (kiocb->ki_obj.user == iocb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1338 | return kiocb; |
| 1339 | } |
| 1340 | return NULL; |
| 1341 | } |
| 1342 | |
| 1343 | /* sys_io_cancel: |
| 1344 | * Attempts to cancel an iocb previously passed to io_submit. If |
| 1345 | * the operation is successfully cancelled, the resulting event is |
| 1346 | * copied into the memory pointed to by result without being placed |
| 1347 | * into the completion queue and 0 is returned. May fail with |
| 1348 | * -EFAULT if any of the data structures pointed to are invalid. |
| 1349 | * May fail with -EINVAL if aio_context specified by ctx_id is |
| 1350 | * invalid. May fail with -EAGAIN if the iocb specified was not |
| 1351 | * cancelled. Will fail with -ENOSYS if not implemented. |
| 1352 | */ |
Heiko Carstens | 002c897 | 2009-01-14 14:14:18 +0100 | [diff] [blame] | 1353 | SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb, |
| 1354 | struct io_event __user *, result) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1355 | { |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 1356 | struct io_event res; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1357 | struct kioctx *ctx; |
| 1358 | struct kiocb *kiocb; |
| 1359 | u32 key; |
| 1360 | int ret; |
| 1361 | |
| 1362 | ret = get_user(key, &iocb->aio_key); |
| 1363 | if (unlikely(ret)) |
| 1364 | return -EFAULT; |
| 1365 | |
| 1366 | ctx = lookup_ioctx(ctx_id); |
| 1367 | if (unlikely(!ctx)) |
| 1368 | return -EINVAL; |
| 1369 | |
| 1370 | spin_lock_irq(&ctx->ctx_lock); |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 1371 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1372 | kiocb = lookup_kiocb(ctx, iocb, key); |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 1373 | if (kiocb) |
| 1374 | ret = kiocb_cancel(ctx, kiocb, &res); |
| 1375 | else |
| 1376 | ret = -EINVAL; |
| 1377 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1378 | spin_unlock_irq(&ctx->ctx_lock); |
| 1379 | |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 1380 | if (!ret) { |
| 1381 | /* Cancellation succeeded -- copy the result |
| 1382 | * into the user's buffer. |
| 1383 | */ |
| 1384 | if (copy_to_user(result, &res, sizeof(res))) |
| 1385 | ret = -EFAULT; |
| 1386 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1387 | |
| 1388 | put_ioctx(ctx); |
| 1389 | |
| 1390 | return ret; |
| 1391 | } |
| 1392 | |
| 1393 | /* io_getevents: |
| 1394 | * Attempts to read at least min_nr events and up to nr events from |
Satoru Takeuchi | 642b512 | 2010-08-05 11:23:11 -0700 | [diff] [blame] | 1395 | * the completion queue for the aio_context specified by ctx_id. If |
| 1396 | * it succeeds, the number of read events is returned. May fail with |
| 1397 | * -EINVAL if ctx_id is invalid, if min_nr is out of range, if nr is |
| 1398 | * out of range, if timeout is out of range. May fail with -EFAULT |
| 1399 | * if any of the memory specified is invalid. May return 0 or |
| 1400 | * < min_nr if the timeout specified by timeout has elapsed |
| 1401 | * before sufficient events are available, where timeout == NULL |
| 1402 | * specifies an infinite timeout. Note that the timeout pointed to by |
Jeff Moyer | 6900807 | 2013-05-24 15:55:24 -0700 | [diff] [blame] | 1403 | * timeout is relative. Will fail with -ENOSYS if not implemented. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1404 | */ |
Heiko Carstens | 002c897 | 2009-01-14 14:14:18 +0100 | [diff] [blame] | 1405 | SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id, |
| 1406 | long, min_nr, |
| 1407 | long, nr, |
| 1408 | struct io_event __user *, events, |
| 1409 | struct timespec __user *, timeout) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1410 | { |
| 1411 | struct kioctx *ioctx = lookup_ioctx(ctx_id); |
| 1412 | long ret = -EINVAL; |
| 1413 | |
| 1414 | if (likely(ioctx)) { |
Namhyung Kim | 2e41025 | 2011-01-12 17:01:08 -0800 | [diff] [blame] | 1415 | if (likely(min_nr <= nr && min_nr >= 0)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1416 | ret = read_events(ioctx, min_nr, nr, events, timeout); |
| 1417 | put_ioctx(ioctx); |
| 1418 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1419 | return ret; |
| 1420 | } |